cassini.c

来自「linux 内核源代码」· C语言 代码 · 共 2,335 行 · 第 1/5 页

/* cassini.c: Sun Microsystems Cassini(+) ethernet driver.
 *
 * Copyright (C) 2004 Sun Microsystems Inc.
 * Copyright (C) 2003 Adrian Sun ([e-mail])
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation; either version 2 of the
 * License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
 * 02111-1307, USA.
 *
 * This driver uses the sungem driver (c) David Miller
 * ([e-mail]) as its basis.
 *
 * The cassini chip has a number of features that distinguish it from
 * the gem chip:
 *  4 transmit descriptor rings that are used for either QoS (VLAN) or
 *      load balancing (non-VLAN mode)
 *  batching of multiple packets
 *  multiple CPU dispatching
 *  page-based RX descriptor engine with separate completion rings
 *  Gigabit support (GMII and PCS interface)
 *  MIF link up/down detection works
 *
 * RX is handled by page sized buffers that are attached as fragments to
 * the skb. here's what's done:
 *  -- driver allocates pages at a time and keeps reference counts
 *     on them.
 *  -- the upper protocol layers assume that the header is in the skb
 *     itself. as a result, cassini will copy a small amount (64 bytes)
 *     to make them happy.
 *  -- driver appends the rest of the data pages as frags to skbuffs
 *     and increments the reference count
 *  -- on page reclamation, the driver swaps the page with a spare page.
 *     if that page is still in use, it frees its reference to that page,
 *     and allocates a new page for use. otherwise, it just recycles the
 *     the page.
 *
 * NOTE: cassini can parse the header. however, it's not worth it
 *       as long as the network stack requires a header copy.
 *
 * TX has 4 queues. currently these queues are used in a round-robin
 * fashion for load balancing. They can also be used for QoS. for that
 * to work, however, QoS information needs to be exposed down to the driver
 * level so that subqueues get targetted to particular transmit rings.
 * alternatively, the queues can be configured via use of the all-purpose
 * ioctl.
 *
 * RX DATA: the rx completion ring has all the info, but the rx desc
 * ring has all of the data. RX can conceivably come in under multiple
 * interrupts, but the INT# assignment needs to be set up properly by
 * the BIOS and conveyed to the driver. PCI BIOSes don't know how to do
 * that. also, the two descriptor rings are designed to distinguish between
 * encrypted and non-encrypted packets, but we use them for buffering
 * instead.
 *
 * by default, the selective clear mask is set up to process rx packets.
 */


#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/compiler.h>
#include <linux/slab.h>
#include <linux/delay.h>
#include <linux/init.h>
#include <linux/ioport.h>
#include <linux/pci.h>
#include <linux/mm.h>
#include <linux/highmem.h>
#include <linux/list.h>
#include <linux/dma-mapping.h>

#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/skbuff.h>
#include <linux/ethtool.h>
#include <linux/crc32.h>
#include <linux/random.h>
#include <linux/mii.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/mutex.h>

#include <net/checksum.h>

#include <asm/atomic.h>
#include <asm/system.h>
#include <asm/io.h>
#include <asm/byteorder.h>
#include <asm/uaccess.h>

#define cas_page_map(x)      kmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
#define cas_page_unmap(x)    kunmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
#define CAS_NCPUS            num_online_cpus()

#if defined(CONFIG_CASSINI_NAPI) && defined(HAVE_NETDEV_POLL)
#define USE_NAPI
#define cas_skb_release(x)  netif_receive_skb(x)
#else
#define cas_skb_release(x)  netif_rx(x)
#endif

/* select which firmware to use */
#define USE_HP_WORKAROUND
#define HP_WORKAROUND_DEFAULT /* select which firmware to use as default */
#define CAS_HP_ALT_FIRMWARE   cas_prog_null /* alternate firmware */

#include "cassini.h"

#define USE_TX_COMPWB      /* use completion writeback registers */
#define USE_CSMA_CD_PROTO  /* standard CSMA/CD */
#define USE_RX_BLANK       /* hw interrupt mitigation */
#undef USE_ENTROPY_DEV     /* don't test for entropy device */

/* NOTE: these aren't useable unless PCI interrupts can be assigned.
 * also, we need to make cp->lock finer-grained.
 */
#undef  USE_PCI_INTB
#undef  USE_PCI_INTC
#undef  USE_PCI_INTD
#undef  USE_QOS

#undef  USE_VPD_DEBUG       /* debug vpd information if defined */

/* rx processing options */
#define USE_PAGE_ORDER      /* specify to allocate large rx pages */
#define RX_DONT_BATCH  0    /* if 1, don't batch flows */
#define RX_COPY_ALWAYS 0    /* if 0, use frags */
#define RX_COPY_MIN    64   /* copy a little to make upper layers happy */
#undef  RX_COUNT_BUFFERS    /* define to calculate RX buffer stats */

#define DRV_MODULE_NAME		"cassini"
#define PFX DRV_MODULE_NAME	": "
#define DRV_MODULE_VERSION	"1.5"
#define DRV_MODULE_RELDATE	"4 Jan 2008"

#define CAS_DEF_MSG_ENABLE	  \
	(NETIF_MSG_DRV		| \
	 NETIF_MSG_PROBE	| \
	 NETIF_MSG_LINK		| \
	 NETIF_MSG_TIMER	| \
	 NETIF_MSG_IFDOWN	| \
	 NETIF_MSG_IFUP		| \
	 NETIF_MSG_RX_ERR	| \
	 NETIF_MSG_TX_ERR)

/* length of time before we decide the hardware is borked,
 * and dev->tx_timeout() should be called to fix the problem
 */
#define CAS_TX_TIMEOUT			(HZ)
#define CAS_LINK_TIMEOUT                (22*HZ/10)
#define CAS_LINK_FAST_TIMEOUT           (1)

/* timeout values for state changing. these specify the number
 * of 10us delays to be used before giving up.
 */
#define STOP_TRIES_PHY 1000
#define STOP_TRIES     5000

/* specify a minimum frame size to deal with some fifo issues
 * max mtu == 2 * page size - ethernet header - 64 - swivel =
 *            2 * page_size - 0x50
 */
#define CAS_MIN_FRAME			97
#define CAS_1000MB_MIN_FRAME            255
#define CAS_MIN_MTU                     60
#define CAS_MAX_MTU                     min(((cp->page_size << 1) - 0x50), 9000)

#if 1
/*
 * Eliminate these and use separate atomic counters for each, to
 * avoid a race condition.
 */
#else
#define CAS_RESET_MTU                   1
#define CAS_RESET_ALL                   2
#define CAS_RESET_SPARE                 3
#endif

static char version[] __devinitdata =
	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";

static int cassini_debug = -1;	/* -1 == use CAS_DEF_MSG_ENABLE as value */
static int link_mode;

MODULE_AUTHOR("Adrian Sun ([e-mail])");
MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
MODULE_LICENSE("GPL");
module_param(cassini_debug, int, 0);
MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
module_param(link_mode, int, 0);
MODULE_PARM_DESC(link_mode, "default link mode");

/*
 * Work around for a PCS bug in which the link goes down due to the chip
 * being confused and never showing a link status of "up."
 */
#define DEFAULT_LINKDOWN_TIMEOUT 5
/*
 * Value in seconds, for user input.
 */
static int linkdown_timeout = DEFAULT_LINKDOWN_TIMEOUT;
module_param(linkdown_timeout, int, 0);
MODULE_PARM_DESC(linkdown_timeout,
"min reset interval in sec. for PCS linkdown issue; disabled if not positive");

/*
 * value in 'ticks' (units used by jiffies). Set when we init the
 * module because 'HZ' in actually a function call on some flavors of
 * Linux.  This will default to DEFAULT_LINKDOWN_TIMEOUT * HZ.
 */
static int link_transition_timeout;



static u16 link_modes[] __devinitdata = {
	BMCR_ANENABLE,			 /* 0 : autoneg */
	0,				 /* 1 : 10bt half duplex */
	BMCR_SPEED100,			 /* 2 : 100bt half duplex */
	BMCR_FULLDPLX,			 /* 3 : 10bt full duplex */
	BMCR_SPEED100|BMCR_FULLDPLX,	 /* 4 : 100bt full duplex */
	CAS_BMCR_SPEED1000|BMCR_FULLDPLX /* 5 : 1000bt full duplex */
};

static struct pci_device_id cas_pci_tbl[] __devinitdata = {
	{ PCI_VENDOR_ID_SUN, PCI_DEVICE_ID_SUN_CASSINI,
	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
	{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SATURN,
	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
	{ 0, }
};

MODULE_DEVICE_TABLE(pci, cas_pci_tbl);

static void cas_set_link_modes(struct cas *cp);

static inline void cas_lock_tx(struct cas *cp)
{
	int i;

	for (i = 0; i < N_TX_RINGS; i++)
		spin_lock(&cp->tx_lock[i]);
}

static inline void cas_lock_all(struct cas *cp)
{
	spin_lock_irq(&cp->lock);
	cas_lock_tx(cp);
}

/* WTZ: QA was finding deadlock problems with the previous
 * versions after long test runs with multiple cards per machine.
 * See if replacing cas_lock_all with safer versions helps. The
 * symptoms QA is reporting match those we'd expect if interrupts
 * aren't being properly restored, and we fixed a previous deadlock
 * with similar symptoms by using save/restore versions in other
 * places.
 */
#define cas_lock_all_save(cp, flags) \
do { \
	struct cas *xxxcp = (cp); \
	spin_lock_irqsave(&xxxcp->lock, flags); \
	cas_lock_tx(xxxcp); \
} while (0)

static inline void cas_unlock_tx(struct cas *cp)
{
	int i;

	for (i = N_TX_RINGS; i > 0; i--)
		spin_unlock(&cp->tx_lock[i - 1]);
}

static inline void cas_unlock_all(struct cas *cp)
{
	cas_unlock_tx(cp);
	spin_unlock_irq(&cp->lock);
}

#define cas_unlock_all_restore(cp, flags) \
do { \
	struct cas *xxxcp = (cp); \
	cas_unlock_tx(xxxcp); \
	spin_unlock_irqrestore(&xxxcp->lock, flags); \
} while (0)

static void cas_disable_irq(struct cas *cp, const int ring)
{
	/* Make sure we won't get any more interrupts */
	if (ring == 0) {
		writel(0xFFFFFFFF, cp->regs + REG_INTR_MASK);
		return;
	}

	/* disable completion interrupts and selectively mask */
	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
		switch (ring) {
#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
#ifdef USE_PCI_INTB
		case 1:
#endif
#ifdef USE_PCI_INTC
		case 2:
#endif
#ifdef USE_PCI_INTD
		case 3:
#endif
			writel(INTRN_MASK_CLEAR_ALL | INTRN_MASK_RX_EN,
			       cp->regs + REG_PLUS_INTRN_MASK(ring));
			break;
#endif
		default:
			writel(INTRN_MASK_CLEAR_ALL, cp->regs +
			       REG_PLUS_INTRN_MASK(ring));
			break;
		}
	}
}

static inline void cas_mask_intr(struct cas *cp)
{
	int i;

	for (i = 0; i < N_RX_COMP_RINGS; i++)
		cas_disable_irq(cp, i);
}

static void cas_enable_irq(struct cas *cp, const int ring)
{
	if (ring == 0) { /* all but TX_DONE */
		writel(INTR_TX_DONE, cp->regs + REG_INTR_MASK);
		return;
	}

	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
		switch (ring) {
#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
#ifdef USE_PCI_INTB
		case 1:
#endif
#ifdef USE_PCI_INTC
		case 2:
#endif
#ifdef USE_PCI_INTD
		case 3:
#endif
			writel(INTRN_MASK_RX_EN, cp->regs +
			       REG_PLUS_INTRN_MASK(ring));
			break;
#endif
		default:
			break;
		}
	}
}

static inline void cas_unmask_intr(struct cas *cp)
{
	int i;

	for (i = 0; i < N_RX_COMP_RINGS; i++)
		cas_enable_irq(cp, i);
}

static inline void cas_entropy_gather(struct cas *cp)
{
#ifdef USE_ENTROPY_DEV
	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
		return;

	batch_entropy_store(readl(cp->regs + REG_ENTROPY_IV),
			    readl(cp->regs + REG_ENTROPY_IV),
			    sizeof(uint64_t)*8);
#endif
}

static inline void cas_entropy_reset(struct cas *cp)
{
#ifdef USE_ENTROPY_DEV
	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
		return;

	writel(BIM_LOCAL_DEV_PAD | BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_EXT,
	       cp->regs + REG_BIM_LOCAL_DEV_EN);
	writeb(ENTROPY_RESET_STC_MODE, cp->regs + REG_ENTROPY_RESET);
	writeb(0x55, cp->regs + REG_ENTROPY_RAND_REG);

	/* if we read back 0x0, we don't have an entropy device */
	if (readb(cp->regs + REG_ENTROPY_RAND_REG) == 0)
		cp->cas_flags &= ~CAS_FLAG_ENTROPY_DEV;
#endif
}

/* access to the phy. the following assumes that we've initialized the MIF to
 * be in frame rather than bit-bang mode
 */
static u16 cas_phy_read(struct cas *cp, int reg)
{
	u32 cmd;
	int limit = STOP_TRIES_PHY;

	cmd = MIF_FRAME_ST | MIF_FRAME_OP_READ;
	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
	cmd |= MIF_FRAME_TURN_AROUND_MSB;
	writel(cmd, cp->regs + REG_MIF_FRAME);

	/* poll for completion */
	while (limit-- > 0) {
		udelay(10);
		cmd = readl(cp->regs + REG_MIF_FRAME);
		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
			return (cmd & MIF_FRAME_DATA_MASK);
	}
	return 0xFFFF; /* -1 */
}

static int cas_phy_write(struct cas *cp, int reg, u16 val)
{
	int limit = STOP_TRIES_PHY;
	u32 cmd;

	cmd = MIF_FRAME_ST | MIF_FRAME_OP_WRITE;
	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
	cmd |= MIF_FRAME_TURN_AROUND_MSB;
	cmd |= val & MIF_FRAME_DATA_MASK;
	writel(cmd, cp->regs + REG_MIF_FRAME);

	/* poll for completion */
	while (limit-- > 0) {
		udelay(10);
		cmd = readl(cp->regs + REG_MIF_FRAME);
		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
			return 0;
	}
	return -1;
}

static void cas_phy_powerup(struct cas *cp)
{
	u16 ctl = cas_phy_read(cp, MII_BMCR);

	if ((ctl & BMCR_PDOWN) == 0)
		return;
	ctl &= ~BMCR_PDOWN;
	cas_phy_write(cp, MII_BMCR, ctl);
}

static void cas_phy_powerdown(struct cas *cp)
{
	u16 ctl = cas_phy_read(cp, MII_BMCR);

	if (ctl & BMCR_PDOWN)
		return;
	ctl |= BMCR_PDOWN;